<div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><DIV>right, except loop is large<BR></DIV><PRE><BR>At 2013-11-07 15:38:25,yuvaraj@multicorewareinc.com wrote:
># HG changeset patch
># User Yuvaraj Venkatesh <yuvaraj@multicorewareinc.com>
># Date 1383809838 -19800
># Thu Nov 07 13:07:18 2013 +0530
># Node ID a8b449fb406d60fa674cfc71d7207bf0d2162175
># Parent 35e1a478048f9fd06b73c461cdea2d7abfe83673
>asm: assembly code for pixel_sad_x4_48x64
>
>diff -r 35e1a478048f -r a8b449fb406d source/common/x86/asm-primitives.cpp
>--- a/source/common/x86/asm-primitives.cpp Thu Nov 07 12:25:14 2013 +0530
>+++ b/source/common/x86/asm-primitives.cpp Thu Nov 07 13:07:18 2013 +0530
>@@ -415,6 +415,7 @@
> p.sad_x4[LUMA_32x32] = x265_pixel_sad_x4_32x32_ssse3;
> p.sad_x4[LUMA_32x64] = x265_pixel_sad_x4_32x64_ssse3;
> p.sad_x3[LUMA_48x64] = x265_pixel_sad_x3_48x64_ssse3;
>+ p.sad_x4[LUMA_48x64] = x265_pixel_sad_x4_48x64_ssse3;
>
> p.luma_hvpp[LUMA_8x8] = x265_interp_8tap_hv_pp_8x8_ssse3;
> p.ipfilter_sp[FILTER_V_S_P_8] = x265_interp_8tap_v_sp_ssse3;
>@@ -470,6 +471,7 @@
> p.sad_x4[LUMA_32x32] = x265_pixel_sad_x4_32x32_avx;
> p.sad_x4[LUMA_32x64] = x265_pixel_sad_x4_32x64_avx;
> p.sad_x3[LUMA_48x64] = x265_pixel_sad_x3_48x64_avx;
>+ p.sad_x4[LUMA_48x64] = x265_pixel_sad_x4_48x64_avx;
> }
> if (cpuMask & X265_CPU_XOP)
> {
>diff -r 35e1a478048f -r a8b449fb406d source/common/x86/sad-a.asm
>--- a/source/common/x86/sad-a.asm Thu Nov 07 12:25:14 2013 +0530
>+++ b/source/common/x86/sad-a.asm Thu Nov 07 13:07:18 2013 +0530
>@@ -2364,6 +2364,178 @@
> lea r3, [r3 + r4 * 2]
> %endmacro
>
>+%macro SAD_X4_48x4 0
>+ mova m4, [r0]
>+ mova m5, [r0 + 16]
>+ mova m6, [r0 + 32]
>+ movu m7, [r1]
>+ psadbw m7, m4
>+ paddd m0, m7
>+ movu m7, [r1 + 16]
>+ psadbw m7, m5
>+ paddd m0, m7
>+ movu m7, [r1 + 32]
>+ psadbw m7, m6
>+ paddd m0, m7
>+ movu m7, [r2]
>+ psadbw m7, m4
>+ paddd m1, m7
>+ movu m7, [r2 + 16]
>+ psadbw m7, m5
>+ paddd m1, m7
>+ movu m7, [r2 + 32]
>+ psadbw m7, m6
>+ paddd m1, m7
>+ movu m7, [r3]
>+ psadbw m7, m4
>+ paddd m2, m7
>+ movu m7, [r3 + 16]
>+ psadbw m7, m5
>+ paddd m2, m7
>+ movu m7, [r3 + 32]
>+ psadbw m7, m6
>+ paddd m2, m7
>+ movu m7, [r4]
>+ psadbw m7, m4
>+ paddd m3, m7
>+ movu m7, [r4 + 16]
>+ psadbw m7, m5
>+ paddd m3, m7
>+ movu m7, [r4 + 32]
>+ psadbw m7, m6
>+ paddd m3, m7
>+
>+ mova m4, [r0 + FENC_STRIDE]
>+ mova m5, [r0 + 16 + FENC_STRIDE]
>+ mova m6, [r0 + 32 + FENC_STRIDE]
>+ movu m7, [r1 + r5]
>+ psadbw m7, m4
>+ paddd m0, m7
>+ movu m7, [r1 + 16 + r5]
>+ psadbw m7, m5
>+ paddd m0, m7
>+ movu m7, [r1 + 32 + r5]
>+ psadbw m7, m6
>+ paddd m0, m7
>+ movu m7, [r2 + r5]
>+ psadbw m7, m4
>+ paddd m1, m7
>+ movu m7, [r2 + 16 + r5]
>+ psadbw m7, m5
>+ paddd m1, m7
>+ movu m7, [r2 + 32 + r5]
>+ psadbw m7, m6
>+ paddd m1, m7
>+ movu m7, [r3 + r5]
>+ psadbw m7, m4
>+ paddd m2, m7
>+ movu m7, [r3 + 16 + r5]
>+ psadbw m7, m5
>+ paddd m2, m7
>+ movu m7, [r3 + 32 + r5]
>+ psadbw m7, m6
>+ paddd m2, m7
>+ movu m7, [r4 + r5]
>+ psadbw m7, m4
>+ paddd m3, m7
>+ movu m7, [r4 + 16 + r5]
>+ psadbw m7, m5
>+ paddd m3, m7
>+ movu m7, [r4 + 32 + r5]
>+ psadbw m7, m6
>+ paddd m3, m7
>+
>+ mova m4, [r0 + FENC_STRIDE * 2]
>+ mova m5, [r0 + 16 + FENC_STRIDE * 2]
>+ mova m6, [r0 + 32 + FENC_STRIDE * 2]
>+ movu m7, [r1 + r5 * 2]
>+ psadbw m7, m4
>+ paddd m0, m7
>+ movu m7, [r1 + 16 + r5 * 2]
>+ psadbw m7, m5
>+ paddd m0, m7
>+ movu m7, [r1 + 32 + r5 * 2]
>+ psadbw m7, m6
>+ paddd m0, m7
>+ movu m7, [r2 + r5 * 2]
>+ psadbw m7, m4
>+ paddd m1, m7
>+ movu m7, [r2 + 16 + r5 * 2]
>+ psadbw m7, m5
>+ paddd m1, m7
>+ movu m7, [r2 + 32 + r5 * 2]
>+ psadbw m7, m6
>+ paddd m1, m7
>+ movu m7, [r3 + r5 * 2]
>+ psadbw m7, m4
>+ paddd m2, m7
>+ movu m7, [r3 + 16 + r5 * 2]
>+ psadbw m7, m5
>+ paddd m2, m7
>+ movu m7, [r3 + 32 + r5 * 2]
>+ psadbw m7, m6
>+ paddd m2, m7
>+ movu m7, [r4 + r5 * 2]
>+ psadbw m7, m4
>+ paddd m3, m7
>+ movu m7, [r4 + 16 + r5 * 2]
>+ psadbw m7, m5
>+ paddd m3, m7
>+ movu m7, [r4 + 32 + r5 * 2]
>+ psadbw m7, m6
>+ paddd m3, m7
>+
>+ lea r0, [r0 + FENC_STRIDE * 2]
>+ lea r1, [r1 + r5 * 2]
>+ lea r2, [r2 + r5 * 2]
>+ lea r3, [r3 + r5 * 2]
>+ lea r4, [r4 + r5 * 2]
>+ mova m4, [r0 + FENC_STRIDE]
>+ mova m5, [r0 + 16 + FENC_STRIDE]
>+ mova m6, [r0 + 32 + FENC_STRIDE]
>+ movu m7, [r1 + r5]
>+ psadbw m7, m4
>+ paddd m0, m7
>+ movu m7, [r1 + 16 + r5]
>+ psadbw m7, m5
>+ paddd m0, m7
>+ movu m7, [r1 + 32 + r5]
>+ psadbw m7, m6
>+ paddd m0, m7
>+ movu m7, [r2 + r5]
>+ psadbw m7, m4
>+ paddd m1, m7
>+ movu m7, [r2 + 16 + r5]
>+ psadbw m7, m5
>+ paddd m1, m7
>+ movu m7, [r2 + 32 + r5]
>+ psadbw m7, m6
>+ paddd m1, m7
>+ movu m7, [r3 + r5]
>+ psadbw m7, m4
>+ paddd m2, m7
>+ movu m7, [r3 + 16 + r5]
>+ psadbw m7, m5
>+ paddd m2, m7
>+ movu m7, [r3 + 32 + r5]
>+ psadbw m7, m6
>+ paddd m2, m7
>+ movu m7, [r4 + r5]
>+ psadbw m7, m4
>+ paddd m3, m7
>+ movu m7, [r4 + 16 + r5]
>+ psadbw m7, m5
>+ paddd m3, m7
>+ movu m7, [r4 + 32 + r5]
>+ psadbw m7, m6
>+ paddd m3, m7
>+ lea r0, [r0 + FENC_STRIDE * 2]
>+ lea r1, [r1 + r5 * 2]
>+ lea r2, [r2 + r5 * 2]
>+ lea r3, [r3 + r5 * 2]
>+ lea r4, [r4 + r5 * 2]
>+%endmacro
>+
> ;-----------------------------------------------------------------------------
> ; void pixel_sad_x3_16x16( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1,
> ; uint8_t *pix2, intptr_t i_stride, int scores[3] )
>@@ -3153,6 +3325,31 @@
> SAD_X3_END_SSE2 1
> %endmacro
>
>+%macro SAD_X4_W48 0
>+%if ARCH_X86_64 == 1
>+cglobal pixel_sad_x4_48x64, 6, 8, 8
>+%define count r7
>+%else
>+cglobal pixel_sad_x4_48x64, 6, 7, 8, 0-4
>+%define count dword [rsp]
>+%endif
>+ pxor m0, m0
>+ pxor m1, m1
>+ pxor m2, m2
>+ pxor m3, m3
>+ mov count, 64
>+
>+.loop
>+ SAD_X4_48x4
>+ SAD_X4_48x4
>+ SAD_X4_48x4
>+ SAD_X4_48x4
>+
>+ sub count, 16
>+ jnz .loop
>+ SAD_X4_END_SSE2 1
>+%endmacro
>+
> INIT_XMM sse2
> SAD_X_SSE2 3, 16, 16, 7
> SAD_X_SSE2 3, 16, 8, 7
>@@ -3198,6 +3395,7 @@
> SAD_X4_W12
> SAD_X4_W24
> SAD_X4_W32
>+SAD_X4_W48
> SAD_X_SSE2 4, 16, 64, 7
> SAD_X_SSE2 4, 16, 32, 7
> SAD_X_SSE2 4, 16, 16, 7
>@@ -3222,6 +3420,7 @@
> SAD_X4_W12
> SAD_X4_W24
> SAD_X4_W32
>+SAD_X4_W48
> SAD_X_SSE2 4, 16, 64, 7
> SAD_X_SSE2 4, 16, 32, 7
> SAD_X_SSE2 4, 16, 16, 7
>_______________________________________________
>x265-devel mailing list
>x265-devel@videolan.org
>https://mailman.videolan.org/listinfo/x265-devel
</PRE></div>